I'd also love a good OCR model

Burrito•4/19/24, 10:52 PM

Very new to AI in general, following https://developers.cloudflare.com/workers-ai/tutorials/explore-code-generation-using-deepseek-coder-models#fill-in-the-middle-code-completion, how would I go about telling it to fill with only one line of code? I tried various { role: 'system' }{ role: 'system' } prompts telling it to fill only one line and it seems to always ignore the instruction.

AndrewMayne•4/20/24, 11:11 PM

I can't find the pricing for text to image. Has that been announced?

AAndrewMayne I can't find the pricing for text to image. Has that been announced?

Isaac McFadyen•4/21/24, 2:17 AM

Text-to-image models are currently in beta, so no, no pricing has been announced yet

Isaac McFadyen•4/21/24, 2:17 AM

Workers AI models generally have pricing released once they go to General Availability

Bilal Mbarki•4/21/24, 3:56 PM

Hi there I'm needing facial recognition in my project and I'm currently using the face_recognition python library on a cloudinstance. Now That I have descovered workers Ai and started thinkering with it a little I'm wondering if there is a model for facialrecognition ( when I ask general models to perform such task they reply that it's against the policy. can I develop my own model and publish it to the edge?

brs29•4/21/24, 4:06 PM

are any more image to text/vision models coming? i'd love to test em out!

brs29•4/21/24, 4:13 PM

i did try that per it being mentioned above but im getting

"InferenceUpstreamError: InferenceUpstreamError: ERROR 3028: Unknown internal error"

"InferenceUpstreamError: InferenceUpstreamError: ERROR 3028: Unknown internal error"

using the same request as i was with uform-gen2-qwen-500muform-gen2-qwen-500m

brs29•4/21/24, 4:17 PM

LLogan Grasby Try passing max_tokens as an input with a value of ~1000. You'll also need to ap...

brs29•4/21/24, 9:52 PM

nice thanks, i was able to get it working, but i did have to simplify my prompt a bit. im sure this is on your radar, but it would be helpful to specify json output

Original message was deleted

brs29•4/22/24, 1:31 AM

idefics2 looks promising, seems like a good fit here since its small-ish and open source

Quill•4/22/24, 2:38 AM

We need a better whisper. current one sucks

Ssilas I would appreciate if you added the large-v3 model. Base is almost unusable for ...

rob•4/22/24, 2:44 AM

Large preferred

LLogan Grasby Try @cf/openai/whisper-tiny-en . Maybe large shows up tomorrow too...

Quill•4/22/24, 3:08 AM

Is version v3 or v2?

playlogo•4/22/24, 6:57 AM

Are there any plans to increase the cap of prompt characters from 6144 to something higher ?

Quill•4/22/24, 7:28 AM

Can whisper return the duration of the input audio?

Winston•4/22/24, 7:51 AM

any know how to pass the negative prompt for stable difussion ai worker? the doc looks like doesn't support negative prompt?

QQuill Can whisper return the duration of the input audio?

Isaac McFadyen•4/22/24, 12:33 PM

It should be fairly easy to compute yourself?

Isaac McFadyen•4/22/24, 12:33 PM

Number of samples / sample rate.

Isaac McFadyen•4/22/24, 12:33 PM

Ex. 961234 samples at 16000 hz is 60.077125 secs

Isaac McFadyen•4/22/24, 12:34 PM

Whisper definitely can return timestamps but I don't think it returns the total duration

rob•4/22/24, 12:39 PM

Ya just do the calc right

IIsaac McFadyen Whisper definitely can return timestamps but I don't think it returns the total ...

Quill•4/22/24, 12:53 PM

Other providers like Azure STT returned the duration. So It would be better for Whisper to return the calculated duration

QQuill Other providers like Azure STT returned the duration. So It would be better for ...

Isaac McFadyen•4/22/24, 12:55 PM

The whisper model itself doesn't return the duration. I suppose it's possible that Workers AI could calculate and return the duration outside of the model, but they'd be doing nothing different than what you would be doing by doing num samples / sample ratenum samples / sample rate

IIsaac McFadyen The whisper model itself doesn't return the duration. I suppose it's possible th...

Quill•4/22/24, 1:02 PM

Thanks for this solution. Btw should I need some library to get audio samples num?

Chaika•4/22/24, 1:13 PM

the catalog show two different llama-3-8b-instruct authors, but meta-llama doesn't work, Error: No such model @hf/meta-llama/llama-3-8b-instruct or taskError: No such model @hf/meta-llama/llama-3-8b-instruct or task

LLogan Grasby @Quill This will return: `{ "text": "Hello, how are you doing today?", "...

Quill•4/22/24, 1:18 PM

I tried @cf/openai/whisper. the returned vtt seems very inaccurate, here is the example input https://github.com/Azure-Samples/cognitive-services-speech-sdk/raw/master/samples/cpp/windows/console/samples/enrollment_audio_katie.wav

LLogan Grasby Try @cf/openai/whisper-tiny-en . Maybe large shows up tomorrow too...

Quill•4/22/24, 4:08 PM

@cf/openai/whisper-tiny-en is worse than @cf/openai/whisper. When will large-v3 show up? Thanks

Bilal Mbarki•4/22/24, 11:06 PM

can you do facial rekon on workers-ai

Cameron Aaron•4/23/24, 2:43 AM

any idea if/when cloudflare will get the Meta Llama 3 70b Instruct

Cameron Aaron•4/23/24, 2:43 AM

I think it might be one of the few models with a knowlege cut off that will work for my app

Cameron Aaron•4/23/24, 2:43 AM

and I think it will be cheaper than gpt 4 turbo

Cameron Aaron•4/23/24, 2:44 AM

I cant really undertstand the pricing because its very non standard

CCameron Aaron I think it might be one of the few models with a knowlege cut off that will work...

Andy zhou•4/23/24, 4:34 AM

it seems cf only deployed small model

AAndy zhou it seems cf only deployed small model

ddts86•4/23/24, 11:24 AM

yes, I would like bigger models.

rzvdrl•4/23/24, 6:11 PM

Is there a way to use a custom model?

Rrzvdrl Is there a way to use a custom model?

Isaac McFadyen•4/23/24, 6:16 PM

?workers-ai-models

Flare•4/23/24, 6:16 PM

Workers AI currently only supports popular open-source models provided by the Cloudflare team. You cannot currently upload your own models or use a model from HuggingFace. See the documentation for the list of Cloudflare-provided models: https://developers.cloudflare.com/workers-ai/models/

Isaac McFadyen•4/23/24, 6:16 PM

I should probably update that because LoRAs are now possible, just not custom models.

rzvdrl•4/23/24, 7:02 PM

It would be awesome if I could import any hf model

Cameron Aaron•4/24/24, 12:20 AM

I run https://slangtranslator.com/ and rn we use GPT 4 turbo do you know a model that might work better that I can use on AI worker like a LORA? Right now i'm using the AI gateway but I think that AI worker might be cheaper than GPT 4 turbo but I cant tell

Slang Translator - Translate Slang to English

Translate slang into standard English. Understand the meaning and context of popular slang terms.

LLogan Grasby Try @cf/openai/whisper-tiny-en . Maybe large shows up tomorrow too...

Quill•4/24/24, 3:00 AM

How about the process? Thanks

CCameron Aaron I run https://slangtranslator.com/ and rn we use GPT 4 turbo do you know a model...

chand1012•4/24/24, 5:40 PM

Llama 3 8B is smarter than some of the early iterations of GPT-4

I'd also love a good OCR model

Similar Threads

Similar Threads

Similar Threads